Advancing entrepreneurship as a design science: developing additional design principles for effectuation

To initialize solutions, genetic algorithms begin with a population of building blocks (also called “solutions” or “means”) (Davis et al. 2007; Goldberg 1989; Holland 1975). Building blocks are the smallest elements that can be mutated or exchanged. Building blocks in entrepreneurship might refer to the parts of a marketing channel or the elements of a business model. The genetic analogue for a building block is a chromosome or meme, a discrete package of knowledge in the form of concepts (Henrich et al. 2008; Heusinkveld et al. 2013), that is either current or imagined. Certain combinations of building blocks constitute organizations that interact with the environment, just as particular combinations of genes form phenotypic expressions that define the biological organisms that interact with the natural world. While an entrepreneur typically has a limited set of means at hand (building blocks), he or she also has a much larger population of means that compete for attention in his or her mind. The set of means the entrepreneur uses to make decisions includes both these real and imagined means (see Dolmans et al. 2014). The initial population of building blocks can be created randomly or by leveraging the prior knowledge of potentially helpful building blocks.

The population of building blocks undergoes evaluation, such that good building blocks are selected. The selection occurs stochastically, based on the building blocks’ relative fitness for the context or purpose. For example, a particular relationship with a supplier as an element in a business model might be selected because it fits the industrial or market context. Selection mechanisms could involve mathematical models, simulations, human heuristics (i.e., let humans intuitively choose building blocks), or some combination thereof.

Variation often involves mutating building blocks or combinations of building blocks. Existing combinations are decomposed into the constitutive building blocks and then recombined to create new combinations (Holland 1975). In genetic algorithm terms, recombination selects genes (i.e., building blocks) from parent chromosomes (i.e., the current combination of building blocks) and creates new offspring. In addition to recombination, mutations take place. Mutation randomly changes the new offspring, and this helps prevent solutions from rendering a local optimum.

Retention is the process by which the new population of building blocks and the combinations they constitute—resulting from selection and variation—replaces the original population. According to evolutionary theory, venturing endeavors or effectual experimentations are bundles of ideas and practices, retained in evolutionary processes, such that those that are not retained do not survive.

All four steps (initialization, selection, variation, and retention) can be done by humans (Kosorukoff 2001). In such human-based genetic algorithms, humans can use their creativity, judgment, and intuition to create, alter, and select solutions without spelling out their decision criteria. Human-based genetic algorithms have designed among others advertisements, music, and architecture, by letting humans contribute directly in any of the four steps (Cheng and Kosorukoff 2004; Ren et al. 2014; Takagi 2001).

Based on these four elementary steps, researchers, in decades of works on genetic algorithms, have identified important design principles (Goldberg 2000, 2002; Holland 1975). Goldberg (2002) cites seven key design principles for genetic algorithms:

This procedure, guided by these design principles, iterates from selection to retention until reaching a terminating condition. The procedure grants successful variations a higher likelihood to survive and forms the basis for future variations, in an evolutionary manner (Andreoni and Miller 1995; Arifovic and Bullard 1997), such that the genetic algorithm gradually approaches satisfying solutions. Solutions are only satisfying because genetic algorithms perform guided but still random search processes, which cannot guarantee optimality of the solutions.

Genetic algorithms are particularly applicable for issues of organization design, as descriptors of how organizations and organizational forms get created, manipulated, and selected (Rivkin and Siggelkow 2003) and how organizations learn through experimentation and learning-by-doing (Davis et al. 2007). Many software companies use genetic algorithms to introduce variations gradually and deemphasize less successful lines of code. Similarly, in product development, genetic algorithms have been applied—for example, by Nokia—to introduce and generate new variants and combinations to make new products and to select iteratively the best-performing variants (Kreng and Lee 2004). Genetic algorithms also are used in a range of other design phenomena in management, such as the emergence of strategic groups (Lee et al. 2002) or the bargaining strategies of negotiators with complete or incomplete information (Zott 2002). For example, in organization design, Bruderer and Singh (1996) used genetic algorithms to explain that managers do not directly predict or design organizational forms but rather guide the evolutionary processes, such as by setting the rate of variation and the selection criteria.

From a genetic algorithm perspective, entrepreneurship is primarily a process in which variation is the rule rather than the exception (Aldrich and Martinez 2001). Entrepreneurs choose the initial pool of building blocks of ideas, resources, and processes in their new venture creations, determine how to increase or decrease variation, and devise how to make selection and elimination decisions (Crawford and Kreiser 2015; McKelvey 2004). Such design decisions can be guided by design principles to aid entrepreneurs in better designing new ventures, and these (additional) design principles can be borrowed from other design fields.

The effectuation process begins with the first principle of effectuation of “who you are, what you know, and whom you know” (Sarasvathy 2001, p. 258), resembling the means at hand. Starting with these means a general aspiration is formed that takes shape through co-creation with committed stakeholders (the third effectuation “crazy quilt” principle) resulting in the convergence of ideas and the emergence of more specific goals (Sarasvathy and Dew 2005). Yet, as entrepreneurs often have a repertoire of contacts and knowledge, they must discern which means serve as good starting points, among all means they might procure (Fisher 2012). Although effectuation theory highlights the key role of means, it does not specify the composition of the set of means. Still, entrepreneurs evaluate alternative means and adjust their means in relation to environmental constraints (Sarasvathy 2001; Sarasvathy and Dew 2005). Therefore, guidance on what means would be good to have in the first place may be helpful (cf. Arend et al. 2015; Chiles et al. 2008; Read and Dolmans 2012).

Viewed through the lens of genetic algorithms, effectuation begins with the building blocks of which entrepreneurs have control (Sarasvathy (2003) also uses the term of building blocks). The key building blocks include knowledge and experience-related elements, such as ideas, beliefs, assumptions, values, interpretative schema, and know-how, as well as available and potential relationships and pre-commitments (for a review of potential building blocks or “units of selection,” see Breslin 2008). The building blocks of earlier start-ups and other career experiences (Engel et al. 2017) are passed on and recombined through the experience and knowledge of founders, employees, customers, investors, and other stakeholders.

Building blocks, therefore, are critical design elements. Yet, in both genetic algorithms and effectuation, designers cannot predict how good the building blocks will perform because of the uncertainty about both the environment and the fit with the building blocks. There is no deterministic landscape against which an effectual designer can pitch the fitness of the (combinations of) building blocks. At the same time, the initial set of means and pre-commitments in combination with the effectuator’s aspiration give a rough direction to the venture; this aspiration also provides initial coherency in the collection of building blocks. The unpredictability of the environment makes it impossible to pick building blocks deterministically. In such a situation, the genetic algorithm instead suggests creating a diverse population of building blocks (Goldberg 2002). Some of the most diverse building blocks in the population will undoubtedly fail, but they may contribute no less to the ultimate satisfying solutions by creating crossovers. Emphasizing effectuation notions of experimentation and failure, the creation of variety adds to the tendency in effectual thinking to go with existing, comfortable means and relationships.

Good building blocks are not static but rather serve as dynamic seeds for creating new and better building blocks, similar to biological seeds that easily break themselves to become plants. Following this logic, effectual “crazy quilting” with partners and stakeholders’ pre-commitments serves as a good design principle to seed building blocks. In line with effectuation theory, the set of building blocks needs to be developed through trusted partnerships, which will practically limit the set of building blocks to what someone can handle in terms of collaborations. Some building blocks are more complementary than others. Good building blocks may evolve into valuable parts of business models and ventures, even if they do not pay off immediately (Sarasvathy et al. 2008). A genetic algorithm-based design process can begin without a minimum standard for any building block (Chattoe-Brown 1998); rather, this minimum standard refers to building blocks’ diversity. As Dew et al. (2018) observed, expert entrepreneurs, in collaboration with their stakeholders, create more and more novel variations than experienced managers. Thus, based on insights from genetic algorithms as well as the findings of Dew et al. (2018), in effectuation, the variety of means might be more important than the sheer number of means.

Returning to our hypothetical example, Myra can benefit from starting with a varied set of means from her and her stakeholders: reaching out and experimenting with different contacts (friends who love to cook, investors, other restaurant owners), gaining a little experience with other restaurants that would lend her a hand, and using her personality to attract clients. Starting with such a diverse set of means is likely better than venturing with a set minimum standard for one particular means (e.g., find a restaurant job to learn for a predefined number of years).

An easy way to increase the diversity of building blocks is to increase the population size of those building blocks. According to the genetic algorithm, when populations are too small, they can lead to premature convergence and substandard solutions (Goldberg 2002), whereas when populations are too large, they can waste valuable time and resources. For effectuation, insufficient building blocks produce a highly localized search, whereas an excessively large population can result in thinly spread resources, scattered efforts, and the starvation of resources for good building blocks. Thus, we expect an inverted U-shaped relationship between the size of the population of building blocks and design fitness, which is, however, unique for each venture. In the same population, diverse and complementary building blocks work better than overlapping ones. In Myra’s situation, although she has no prior restaurant experience and is not a good cook herself, starting with all her available means would be too much, so the resulting effectual design would suffer. If she began with just one means—for example, a particular friend who loves cooking—she may end up with a restaurant far from the best she could have developed.

Design principle 1: Gather more than just a few means; at the same time, it is more important to have more diverse means than merely more means.

Because building blocks are not static, design principles must be in place to ensure that the better building blocks grow and take over a dominant share of the population in the effectuation process. Whereas in biological evolution selection is mainly attributed to environmental forces (Hodgson 2013), in genetic algorithms and effectuation, the selection is predominantly driven by decision-makers. In organizational settings, building blocks compete for the attention of human beings (Ocasio 1997), implying that humans move on from certain building blocks if they stop noticing their public expressions, internalizing them, or reproducing them (Weeks and Galunic 2003). For example, new ideas or new procedures are put down if they are not remembered and enacted.

In effectuation theory, the entrepreneur is the “pilot” who makes selection decisions in collaboration with committed stakeholders (Sarasvathy 2008). Thus, in this selection, the fitness of any particular means is subjective or intersubjective, not an objective given (Chiles et al. 2008; Dolmans et al. 2014). For effectual entrepreneurs, selection is an internal process of co-adaptation, involving their relevant stakeholders (Dew et al. 2018). Thus, the primary selection mechanisms are the direction set out by working with the set of means in view of a general aspiration and the co-creation processes with stakeholders. Yet, entrepreneurs might need more guidance in making selections, and thus, it is valuable to specify useful selection mechanisms that account for the uncertainty of the situation and lack of a predefined market environment that would determine the right fit.

The sense of whether something works or not in an individual’s mind is explained in psychology by self-regulation theories, in particular control theories (e.g., Eccles and Wigfield 2002; Zimmerman 2002). These theories posit that individual entrepreneurs create a perception of the state of a variable of interest and compare the perception to a referent, which represents the desired state. The discrepancies between these two states determine the (lack of) fitness. The desired state is known in effectuation as a general aspiration (Sarasvathy 2008). In many cases, while it is hard to devise an algorithm for calculating the fitness of a design, it is relatively easy to tell the relative change in fitness (i.e., if a change increases or decreases the fitness). Similarly, Myra has a hard time to pinpoint the exact fitness score of a dish, say curry mutton, but she can tell relatively easily whether changing curry mutton to curry fish pleases more customers, and thus, she would be able to tell whether building blocks are fitness-reducing versus fitness-enhancing. As in genetic algorithms, what matters is the relative performance improvement instead of the exactitude of performance.

More importantly, because it is impossible to define perfect criteria to evaluate and select building blocks, due to dynamic uncertainty (Schmitt et al. 2018) about their future performance, designers of genetic algorithms focus on designing fast-and-frugal tests. Such fast-and-frugal tests are also essential for effectuation (Sarasvathy 2001, 2003) and widely recognized by practitioners, such as in the “lean start-up” movement (Ries 2011). Effectuation has one main design principle to guide selection, namely affordable loss, which prescribes that an entrepreneur should decide on the basis of what he or she can afford to lose (Dew et al. 2009; Sarasvathy 2008). Along this principle, we can identify other principles that guide selection based on genetic algorithm research, for example: running out of time, achieving sufficient quality, substantial convergence in the population (implying a decrease in diversity), and time elapsed since the last improvement.

Responding to calls to include time in the affordable loss principle (Dew et al. 2009), we point at the running out of time criterion which refers to the time someone can afford to test a certain solution, for instance allowing 2 months as an affordable period to validate an idea through customers. Using such rules, Myra tests the idea of working with a “wanna-be-chef” friend for 2 months, at the end of which Myra uses the market’s responses to decide on the idea. Using the principle of achieving sufficient quality, entrepreneurs can similarly create simple rules, such as “Can we find two major buyers for the current service or product in three months?” Can Myra fill 50% of her restaurant tables in half a year? Finally, the time elapsed since last improvement criterion is again based on the time someone can afford, this time to come up with an improvement. It could lead to a simple rule, such as “If we cannot improve a certain technology after working on it for a year, we move on to other technologies” or “If we cannot find another key customer segment in two months, we have to change our approach.” Practicing entrepreneurs, consciously or subconsciously, use such guiding principles, but these have not been recognized as design principles for guiding effectual decisions, nor have they been studied explicitly.

The preferences indicated by these tests do not require global knowledge and are not deterministic in nature (Sastry et al. 2005; Sastry and Goldberg 2003). Instead, these nondeterministic fast-and-frugal selection principles help to maintain a necessary level of diversity under conditions of fundamental uncertainty. These fast-and-frugal tests do not assume existing markets or customer preferences but are low fidelity design principles for situations where entrepreneurs did not figure out their market yet, or in cases where entrepreneurs actually transform markets themselves (e.g., Dew et al. 2011; Sarasvathy and Dew 2005).

Design principle 2: Use one or more fast-and-frugal tests (i.e., substantial convergence in the population, sufficient quality, time elapsed since last improvement) to select appropriate building blocks for new venture creation.

The creation of variation, a crucial process in effectuation, occurs in collaboration with stakeholders, as specified in the third effectuation principle (Sarasvathy 2003, 2008; cf. Dew et al. 2018). As such, variations are not just a blind try but often guided by previous experience in a co-creating process (cf. Alvarez et al. 2013; Engel et al. 2017). New building blocks are introduced along the way by leveraging contingencies (the fourth effectuation principle “lemonade”). Dew et al. (2018) find that expert entrepreneurs create a greater range of variations than experienced managers do. These expert entrepreneurs also create variations based on their previous experience in addition to random variations (Sarasvathy et al. 2010). In a study on the creation of novelty through effectual transformations, Dew et al. (2011) found inductively that expert entrepreneurs mentioned a myriad of processes to introduce variations: deletion and supplementation, composition and decomposition, exaptation, manipulation, deformation, localization, prototyping, stereotyping, and free association. These variation processes, from an evolutionary view, are fundamentally generated by mutation and crossover of building blocks. Crossing over properly requires decomposition (Goldberg 2002), which reduces structural interdependencies and thereby allows for easy evolution (Simon 1969). Thus we emphasize here the process of composition, which is seen as an essential prerequisite of designing high-performing, enduring artifacts in effectuation theory (Sarasvathy 2003, 2008). Without decomposition, variation is limited, as faced by novice entrepreneurs (Sarasvathy et al. 2010). However, the role of iterative and continued decomposition is less obvious to nondesigners, remains underspecified (Sarasvathy and Dew 2005), and is often neglected (Sarasvathy 2008), so we elaborate specifically on the process of decomposition in effectuation.

Decomposition helps in two main ways. First, it helps to identify building blocks that can be changed, reused, and recombined, as described by the logic of exaptation that existing elements get co-opted for new roles (Dew et al. 2004). Different decompositions will surface distinct building blocks and thus enable distinct exchanges, leading to distinct exaptations. Second, decomposing helps to reveal the relationships and hence the interconnectedness among the building blocks in which the whole is greater than the sum of its parts.

To decompose better, genetic algorithms can provide some guidance. Different decompositions should be considered—as a thought exercise—to identify distinct sets of building blocks, some of which may evolve toward competent solutions better than others (Goldberg 2002). Empirical evidence confirms that good designers are those with the ability to find various representations of a design issue and the multitude of representations help them to solve hard problems (Goldberg 2002). Theoretically, such representations might include economic–financial representations (as found in accounting textbooks), real options theory (Kogut and Kulatilaka 2001; Vassolo and Anand 2008; Zhang and Babovic 2011), and dynamic capability views (Teece et al. 1997). The multitude of representations enable searches for new decompositions; those new decompositions can then help to reveal building blocks that are hard to find, deep, complex, integrated, and difficult to separate. Many capabilities, routines, and imprint effects are tough to reveal because they have developed tacitly, through years of experience. Without attempts to facilitate decomposition, designers can rely only on good luck to find good building blocks.

Myra thus might try a financial decomposition, which helps to reveal the costs, revenues, and margins and which, in the end, helps her realize that starting a conventional restaurant is too expensive. Therefore, she decides to start with an Indian-themed catering service, using the help from her friends who cook at home. Within a few weeks, the business has gained some success, especially catering to office parties that want to try something different. With such clients in mind, Myra drafts a simple flow diagram, drawn from a business model canvas, to decompose her business into its key building blocks: creating the menu, buying food and materials, preparing the food with key people, and serving the food. Finally, Myra asks a friend for input, who helps her decompose the capabilities in her new venture. Adding these different decompositions to her drawing board gives Myra different sets of building blocks that she can change to iterate her new venture. For example, she identifies that the process and capability of buying food and materials can be broken down and improved, and therefore, she engages more actively with a friend who knows a great deal about food and materials.

Design principle 3: Decompose by constructing multiple representations to identify building blocks and subsequently explore new combinations by exchanging building blocks.

Retention means deciding which new means, generated through variation, are kept as the new basis. For example, Myra may try varying her dishes, either intentionally or accidentally, and then decide on whether to update the dish with those variations. Effectuation theory states that the entrepreneur, as the pilot-in-the-plane, makes these updating decisions based on the available information at that time. Genetic algorithms teach us about the speed of updating processes and indicate that the decisions on building blocks should be neither too fast nor too slow. A decision that takes too long would imply a waste of resources on building blocks that fail to produce eventually, likely at the cost of investing in those that are productive. However, weeding out (combinations of) building blocks too quickly fails to provide the time for the building blocks to develop and become productive. Furthermore, building blocks cannot be put away deterministically with respect to the rate of introducing new ones, because if so, the diversity needed to create improvements disappears, and the success rate falls into local improvement. Many seasoned entrepreneurs clearly know the importance of allowing time and patience before making final decisions about the experimentation in which they engaged. Thus, we anticipate an inverted U-shaped relationship between the time taken for selection and design fitness.

As soon as Myra started a restaurant, alongside her catering business, she chose to focus on drawing in discovery-driven clients. She plans to introduce a new dish every other day, and if the new dish does not sell well in those 2 days, she will remove it. In reality though, if a dish does not perform well, Myra often takes weeks to adjust it and changes a dish only if it fails consistently for 2 weeks. The reason for this lengthened elimination decision is that only one chef works at the restaurant, so the speed of generating new dishes is relatively slow. However, to draw in clients, Myra realizes she has a large collection of Indian music and can experiment with different songs; she does not hesitate to stop the song if it fails to appeal to customers, even after just a few seconds. Therefore, she adjusts the speed of her selection decisions to the time required, such as the period needed to develop a new dish.

Yet the speed of updating the building blocks is interdependent with the severity of these decisions and is dependent on the amount of variation created. Thus, selection must align with the rate of introducing changes and variations. Change is valuable for innovation and entrepreneurship, so a higher frequency of change might appear preferable. Moreover, a high mutation rate early on is helpful to get out of local maxima. However, mutation rates also can be too high, as research in both genetic algorithms and biology shows (e.g., Domingo-Calap and Sanjuán 2011; Elena and Sanjuán 2005; Gordo and Sousa 2010). The occurrences of mutation and recombination that increase fitness are far less frequent than those that decrease it. If the rate of change is too high and many changes occur together, entrepreneurs cannot effectively remove underperforming building blocks (i.e., abort and change some of their venturing experimentation) in favor of better ones. Because the detrimental changes cannot be weeded out in time if the rate of change is too high, the selection mechanism needs to be adjusted. In contrast, if the selection is too fast with respect to the rate of change, new ideas are not given enough time to evolve before they are judged, likely leading to biases against new and risky changes. Ventures with such unbalanced mechanisms tend to develop a culture contrary to changes (Denrell and March 2001).

For Myra’s discovery-driven clients, the ambience of her new restaurant, including visual and acoustic effects, is a key design factor. To alter the acoustic effect, Myra simply changes her CDs; to alter the visual effect, she would need to redecorate, which represents a much slower process. Therefore, if a song is not generating its desired effect, Myra quickly pulls out another option from her large collection of Indian music. Conversely, Myra would only change the decor if she becomes totally convinced, from feedback from many clients, of the need for change, because it takes much longer to get the decoration right.

Design principle 4a: Align selectivity with the rate of change and severity of change.

Design principle 4b: Align selection decision speed with the rate of change and severity of change.

Table 2 contains an overview of all the design principles related to the genetic algorithm steps, linking the new and existing effectual principles and illustrating the new ones with examples. In the iterative genetic algorithm-based design process, these design principles become inherently interrelated, consistent, and collectively independent (Romme and Endenburg 2006). That is, the design principles are interrelated because they cooperate iteratively in the four-step evolutionary process (e.g., without decomposition, there is no population of building blocks). Some design principles coordinate with other design principles, such as principle 4.

This study helps substantiate an approach to develop entrepreneurship as a science of the artificial. In fact, as a field, we know little about the approaches to fulfill the promise of developing entrepreneurship as a science of the artificial (cf. Venkataraman et al. 2012). Our approach to develop design principles uses analogous transfer from other design sciences, which inevitably introduces new concepts and vocabularies for entrepreneurship and effectuation scholars. However, these new vocabularies carry the potential to further enrich entrepreneurship and management theory. In performing an analogous transfer from a distant field such as genetic algorithms, we shed light on how future research might similarly cross-fertilize from other design fields to shape the “world as it might be” for entrepreneurship as a science of the artificial (Sarasvathy et al. 2008). In our field, this approach is rarely used, which may be due to the different vocabularies and different (prescriptive vs. positivist) design traditions (Van Burg and Romme 2014). We hope this initial effort opens the door for more cross-pollination for entrepreneurship from other sciences, particularly other evolutionary artificial intelligence design techniques. This approach of borrowing could help to develop not just high-level design principles but also design elements and approaches for entrepreneurship theory and practice.

Design principles from an important link between the scientific body of knowledge and practice (Berglund et al. 2018; Romme 2003), and entrepreneurs can apply the principles we developed to design new ventures. Thus, our study reconnects to Herbert Simon’s notion of design as a science of the artificial concerned with the “world as it might be” (Sarasvathy et al. 2008). Simon’s notion is crucial for advancing entrepreneurship as a design science (Dimov 2016), to prescribe real help for entrepreneurs (Sarasvathy and Venkataraman 2011). In the same vein, this study shows how to develop principles that offer prescriptions and, in doing so, helps bridge the gap between descriptive academic research and entrepreneurial practice.

This article contributes to effectuation theory in two critical ways: its theoretical refinement as a design science (e.g., Arend et al. 2015; Read et al. 2016; Venkataraman et al. 2012) and its provision of useful prescriptions for designing effectuation processes (e.g., Sarasvathy 2001, 2008). Specifically, this article provides new and complementary design principles for effectuation—a critical challenge for effectuation to grow as a design science (Sarasvathy and Dew 2005). Our new design principles pertain to crucial issues in effectuation theory, such as how to deal with variation, how to select means, and at what moment. Such key issues are faced by effectuators and puzzle entrepreneurship scholars, but they have not been specified yet in the initial set of five effectual design principles.

Table 3 illustrates how the identified principles augment extant effectuation principles and enrich effectual design processes. In particular, we provide more guidance on the means that entrepreneurs can use to build their ventures (Arend et al. 2015; Chiles et al. 2008; Read et al. 2016), highlighting the relevance of a diverse yet reasonable amount of means. The newly identified principle 1 reveals that the set of means should be primarily diverse and does not necessary to be a large set. The recognition that the diversity of means is important highlights an overlooked aspect of effectuation, moving effectuation beyond merely starting from means to the question which means represent better starting points.

The differentiation of better means to start with can offer theoretical insights that may explain the lack of consistent results on the relative effects of a means-driven orientation (e.g., Brettel et al. 2012; Read et al. 2009). These inconsistent results may be due to failures to discriminate between well-populated, diverse sets of means and overpopulated sets that lack diversity. Such differences lead to differential performance, such that entrepreneurs with more diverse sets of means achieve better performance than those with an abundance of one particular means, such as capital (e.g., Dolmans et al. 2014). In addition, the new principles 2 and 3 indicate how entrepreneurs can co-adapt their means in an evolutionary process. Together, these principles suggest the diversity and size of the population and selection of means as important avenues for future effectuation research.

Moreover, design principle 4 issues clear prescriptions for moving on from certain means. Prior effectuation research suggests that people tend to continue expanding the set of means (Read and Dolmans 2012) but offers little guidance on how to deal with unsatisfying means. Our newly identified principles provide more guidance on the generation of variety and the selection of variants than the initial set of five principles do. While the initial set of principles generically treats variants as a contingency to exploit and fails to guide on when to move on from certain means, our newly identified principles 3 and 4 offer guidance on the most important issues regarding variation and retention to distinguish what are good means. Moreover, these principles provide guidance regarding an important temporal issue in effectuation (cf. Jiang and Rüling 2018), namely, the speed at which variants need to be introduced and eliminated.

The newly identified design principles address critical but so far often untouched questions that entrepreneurs face in designing their ventures, including which means to start with, how to select the means in relation to their performance over time, how to identify and create new means, and how to balance the pace and selectivity of attaining new means and eliminating subperforming ones. These questions are untouched in the literature to date, but even simple awareness of these design questions would help frame entrepreneurs’ minds and enhances designers’ learning capability, along with the effectiveness of their actions (cf. Romme and Endenburg 2006), and the new principles guide entrepreneurs in their decision-making in several key design questions.

Entrepreneurship, AI, and big data are all relatively new and fast-growing disciplines. The fields of big data and AI deal with huge amounts of information and have focused on developing fast-and-frugal tests, and accordingly, they offer potential lessons to the entrepreneurship field via analogous transfer or cross-pollination. First, genetic algorithms and other evolutionary metaheuristics seem particularly helpful for furthering entrepreneurship theories, both theoretically and practically. For example, big data and AI may shed light on the design principles of bricolage (Baker and Nelson 2005) or theories on opportunity creation and discovery (Alvarez and Barney 2007).

Second, the design principles we identified are high-level design principles, just like the initial set of five effectual principles of Sarasvathy (2001). Therefore, it would be worthwhile to develop additional design principles by analogous transfer from AI around the processes of crossover and mutation, along the lines of the nine forms of transformation as identified by Dew et al. (2011).

Third, the additional principles based on genetic algorithms can be formulated readily as novel and intriguing empirical research questions. For example, does nondeterministic fast-and-frugal test lead to better venture designs than other forms of evaluation do? It is important to note that as in good design, the design principles form an interrelated system and cannot be tested as completely independent propositions, due to the importance of their interdependencies (Caspin-Wagner et al. 2013). A system approach is necessary for the theoretical and empirical development of these design principles. In this respect, an integrative model to simulate the entrepreneurship process would be helpful (Davis et al. 2007; McKelvey 2004).

Fourth, although these principles cohere theoretically in the iterative, four-step genetic algorithm process, the coherency can be further improved by future work that puts the principles into practice (cf. Van Burg et al. 2008). Thus, the coherence of the design principles requires testing, and recent work on measuring ideas in entrepreneurship (e.g., Hill and Birkinshaw 2010) may help in this regard.

Finally, the fields of AIs and big data develop computational models to handle large amount of information nondeterministically, and they can be applied literally as decision support systems for people running new ventures or innovation projects (cf. Dellermann et al. 2018; Sohn and Lee 2013; Zhang and Babovic 2011, 2012). Future studies can explore how entrepreneurs can use AI nondeterministic models as decision support for path-dependent decision-making under uncertainty.